CN103390169A - Sorting method of vehicle-mounted laser scanning point cloud data of urban ground objects - Google Patents

Abstract

The invention provides a method for classifying urban features of vehicle-mounted laser scanning point cloud data, which is mainly divided into two steps: combining the driving trajectory data, performing road surface filtering on the original vehicle-mounted laser scanning point cloud data, and converting the road surface point cloud data It is separated from other feature point cloud data; the obtained other feature point cloud data is further identified as different feature types. In the road surface filtering algorithm, the present invention combines the driving trajectory data and introduces the histogram statistical analysis method, which greatly improves the degree of self-adaptation, automation and accuracy of the road surface filtering algorithm in the urban environment; in other ground object classification algorithms , the invention improves the point cloud feature image classification method based on the maximum elevation value, and uses the cellular automaton technology to correct the misclassified point cloud data, and provides a new idea and method for the vehicle laser scanning point cloud data classification algorithm method.

Description

A Classification Method of Urban Land Objects Based on Vehicle-mounted Laser Scanning Point Cloud Data

technical field

The invention relates to the technical field of vehicle-mounted laser radar, in particular to a method for classifying point cloud data of vehicle-mounted laser scanning.

Background technique

In the 1970s, NASA developed the laser radar measurement technology (LIDAR, Light Detection And Ranging), which provided a new technical means for obtaining spatial information. By the 1980s, the University of Stuttgart in Germany combined lidar technology with a timely positioning and attitude determination system, using aircraft as a carrier, and successfully developed an airborne lidar measurement system (ALSS, Airborne Laser Mapping System). In the late 1980s, the vehicle-mounted mobile measurement system (MMS, Mobile Mapping System) with the vehicle as the platform was successfully developed. This technology provides a reliable, flexible and efficient way to quickly collect high-precision 3D data of roads and related features on both sides of the road. However, the original vehicle-mounted laser scanning point cloud data is actually a collection of massive 3D coordinate points that are disorganized, including various ground objects such as roads, buildings, trees, and poles, and the data volume is large. Therefore, before further processing or 3D modeling of vehicle-mounted laser scanning point cloud data, if it can be accurately classified, the efficiency and accuracy of subsequent data processing will be greatly improved.

Compared with vehicle-mounted laser scanning point cloud data, research on classification methods for airborne laser scanning point cloud data is more mature. For example: "An automatic classification method for airborne laser radar point cloud data" (application number: 200910272643.0), "a fast filtering method for airborne laser radar point cloud data" (application number: 201110337099.0), "A kind of airborne laser Intelligent Filtering Method for Point Cloud Data" (Application No.: 201210350254.7), the three invention patents appealed for corresponding filtering methods and classification methods for airborne laser point cloud data. However, due to the different operation methods, the vehicle-mounted laser scanning point cloud data is quite different from the airborne laser scanning point cloud data, so the classification method of the airborne laser scanning point cloud data cannot be directly applied to the vehicle-mounted laser scanning point cloud data.

In recent years, many scholars at home and abroad have proposed some classification methods for vehicle-mounted laser scanning point cloud data. However, the degree of self-adaptation of these methods is low, and more thresholds are usually required to be manually set, and it is impossible to achieve efficient and accurate classification for vehicle-mounted laser scanning point cloud data in complex urban environments.

Contents of the invention

The purpose of the present invention is to solve the existing problems in the prior art, combining the driving track data of the vehicle-mounted laser scanning system, and introducing technologies such as histogram statistical analysis and cellular automata, to improve the current vehicle-mounted laser scanning point cloud data classification method degree of automation, efficiency and accuracy.

The technical solution provided by the present invention includes a method for classifying urban features of vehicle-mounted laser scanning point cloud data, comprising the following steps:

Step 1, combined with the driving trajectory data, perform road surface filtering on the original vehicle-mounted laser scanning point cloud data, and separate the road surface point cloud data from other object point cloud data, including the following sub-steps,

Step 1.1, according to the driving trajectory data, along the direction of the driving trajectory, with each node of the driving trajectory as the center point, construct a rectangular filtering window, perform histogram statistical analysis on the data falling in the rectangular filtering window to determine the threshold range of the road surface, and obtain a preliminary Road surface point cloud data;

Step 1.2: Analyze the preliminary road surface point cloud data obtained in step 1.1 from the two dimensions of distance and angle, arrange according to the scanning line arrangement, and mark the location of each road surface data point in the preliminary road surface point cloud data. scan line number;

Step 1.3: Analyze the road surface point cloud data obtained after the scanning lines in step 1.2 are scanned line by scan line according to the order of the scan line numbers, including calculating the relationship between each road surface data point in the scan line and its The elevation difference of two adjacent road surface data points, if the two elevation differences exceed the preset elevation difference threshold, then the road surface data point is removed as a noise point; the preliminary road surface point cloud data obtained from step 1.1 is removed After the noise points, the retained road surface data points are used as the final road surface point cloud data, and the final road surface point cloud data is removed from the original vehicle laser scanning point cloud data to obtain other ground object point cloud data;

Step 2, further identifying the point cloud data of other features obtained in step 1 as different types of features, including the following sub-steps,

Step 2.1, take each driving trajectory node as the center point, construct a circular area according to the preset, search for the road surface point cloud data falling in the circular area according to the results obtained in step 1.3, and calculate all the roads in each circular area in turn The average elevation value of the surface data points is used as the road surface elevation value at the corresponding driving trajectory node;

Step 2.2, traverse and calculate the horizontal distance between each data point in the point cloud data of other ground objects and all the driving track nodes, select the driving track node closest to the currently traversed data point, and set the elevation value of the currently traversed data point to Make a difference with the elevation value of the road surface near the corresponding driving track node obtained in step 2.1, and obtain the relative elevation value of each data point in the point cloud data of other features with respect to the nearby road surface;

Step 2.3, according to the preset sampling interval, construct a rectangular regular grid on the horizontal plane, project the point cloud data of other ground objects obtained after the elevation correction in step 2.2 into the regular rectangular grid on the horizontal plane, and count in each grid unit The maximum elevation value of all data points is used as the feature value of the grid cell to generate a point cloud feature image;

In step 2.4, preliminary classification is carried out through the preset category threshold, and the number corresponding to the category of each grid unit is obtained, which constitutes the preliminary classification result of other ground objects;

In step 2.5, the preliminary classification results of other ground objects obtained in step 2.4 are evolved by using the cellular automaton for a preset number of times to obtain the final classification results of other ground features.

Moreover, step 1.1 includes the following sub-steps,

Step 1.1.1, according to the range covered by the original vehicle-mounted laser scanning point cloud data, construct a two-dimensional spatial regular grid index, calculate the number of the grid unit to which each data point belongs and generate an index file;

Step 1.1.2, traversing the driving track nodes, building a rectangular filtering window in turn, and searching for all data points falling in the rectangular filtering window according to the index file;

Step 1.1.3, carry out histogram statistical analysis to the elevation values of all data points in the rectangular filtering window obtained in step 1.1.2, the statistical method is as follows,

Take the elevation of the lowest data point in the current rectangular filtering window as the starting point, partition according to the length of the preset interval from low to high, set it into n intervals, and count the points falling in each interval in turn, and in order from low to high Number the n intervals as 0, 1, 2, ..., n-1; select the 5 intervals with the most points, and determine the interval numbers corresponding to these 5 intervals;

Arrange the 5 interval numbers in ascending order, take the interval MaxBin with the most points as the center, traverse the remaining 4 intervals up and down, and select the intervals adjacent to MaxBin in the remaining 4 intervals, form a continuous interval segment;

Check whether the minimum interval number in the continuous interval segment is 0,

If it is 0, then the continuous interval segment is used as the road surface threshold range, and the data points falling in the continuous interval segment are identified as road surface data points to obtain preliminary road surface point cloud data;

If it is not 0, the continuous interval segment and all interval segments lower than the continuous interval segment are regarded as the road surface threshold range, and the data points in the continuous interval segment and all data points lower than the continuous interval segment are considered is the road surface data point, and the preliminary road surface point cloud data is obtained.

Moreover, the construction rule of the cellular automaton in step 2.5 is that the cell is the grid unit after the preliminary classification in step 2.4, the state value of each cell is the category number of the grid unit, and the cellular space is the grid unit after the step 2.4 A collection of all grid units after preliminary classification;

Each evolution traverses all the cells in the cellular space once according to the set cellular automata rules, and the cellular automata rules include the following sub-steps,

Step a, judge whether the state value S of the current cell is 0, if it is 0, do not make subsequent judgments and keep the state value, if it is not 0, continue to the subsequent step b;

Step b, count the number sum _type of all state values in the Moore neighbors of the current cell, if the number sum _k of a certain state value k is greater than 4, change the state value of the current cell to k and end the The discrimination of the cell, if it does not exist, continue to the subsequent step c;

Step c, count all state values in the Moore neighbors of the current cell to obtain the maximum state value S _max , if S _max is greater than the state value S of the current cell, change the state value of the current cell to S _max , otherwise keep it unchanged.

The technical solution provided by the invention combines road surface filtering and other surface object classification. In the road surface filtering algorithm, the present invention combines the driving trajectory data and introduces the histogram statistical analysis method, which greatly improves the degree of self-adaptation, automation and accuracy of the road surface filtering algorithm in the urban environment; in other ground object classification algorithms , the invention improves the point cloud feature image classification method based on the maximum elevation value, and uses the cellular automaton technology to correct the misclassified point cloud data, and provides a new idea and method for the classification of vehicle-mounted laser scanning point cloud data .

Description of drawings

Fig. 1 is the overall flowchart of the embodiment of the present invention;

Fig. 2 is the flow chart of road surface filtering of the embodiment of the present invention;

Fig. 3 is a schematic diagram of building a rectangular filter window on a road surface according to an embodiment of the present invention;

FIG. 4 is a flow chart of moving window filtering based on driving trajectory data according to an embodiment of the present invention;

FIG. 5 is a flow chart of other object classification in the embodiment of the present invention;

Fig. 6 is a flow chart of preliminary classification of other features based on the point cloud feature image method according to an embodiment of the present invention;

FIG. 7 is a Moore-type neighbor model according to an embodiment of the present invention, that is, an eight-neighborhood neighbor model.

Specific implementation method

The technical solution of the present invention can adopt computer software technology to realize the automatic operation process. In order to facilitate those skilled in the art to understand and implement the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

Referring to Fig. 1, an embodiment of the present invention provides a method for classifying urban features of vehicle-mounted laser scanning point cloud data, including:

Step 1. Combining with the driving track data, perform road surface filtering on the original vehicle laser scanning point cloud data, and separate the road surface point cloud data from other object point cloud data;

In step 2, the point cloud data of other features obtained in step 1 after excluding the point cloud data of the road surface are further identified as different types of features such as buildings, poles, trees and other types of features.

Further as a preferred embodiment, with reference to Figure 2, the step 1 includes:

Step 1.1, moving rectangular filtering window filtering algorithm based on driving trajectory: According to the driving trajectory data, along the direction of driving trajectory, with each driving trajectory node as the center point, construct a rectangular filtering window, and filter the data falling in the moving rectangular filtering window Statistical analysis of the histogram is carried out to determine the threshold range of the road surface and obtain the preliminary road surface point cloud data.

Referring to Fig. 3, the construction method of embodiment rectangular filtering window is specifically:

可以得到垂直

的向量

矩形滤波窗口的相邻两边则分别平行于

方向和

方向。在矩形滤波窗口中，平行于

方向的边长长度length为

平行于

方向的边长长度width只要大于道路面宽度即可，如图中行车轨迹节点O左右各扩展至-width/2、+width/2。具体实施时，本领域技术人员可自行根据具体道路情况预设width取值，本实施例中width取40米。Along the direction of the driving trajectory, a rectangular filtering window is constructed with each driving trajectory node as the center point. For each driving trajectory node O, construct a vector from the driving trajectory node O to its adjacent driving trajectory node A

can get vertical

vector of

The adjacent two sides of the rectangular filtering window are respectively parallel to

direction and

direction. In the rectangular filter window, parallel to

The side length length of the direction is

parallel to

The side length width of the direction only needs to be greater than the width of the road surface, as shown in the figure, the left and right sides of the driving track node O are respectively extended to -width/2 and +width/2. During specific implementation, those skilled in the art can preset the value of width according to specific road conditions. In this embodiment, the width is 40 meters.

Further as a preferred embodiment, with reference to Figure 4, the step 1.1 includes:

Step 1.1.1, according to the range covered by the original vehicle-mounted laser scanning point cloud data, construct a two-dimensional spatial regular grid index. It is recommended to set the grid side length to half of the side length width of the rectangular filter window, that is, 20 meters. Traverse the original vehicle-mounted laser scanning point cloud data, calculate the grid cell numbers i and j to which each data point belongs, and generate an index file. Among them, i represents the grid cell number in the horizontal direction, j represents the grid cell number in the vertical direction, and the value range of i and j depends on the range covered by the original vehicle-mounted laser scanning point cloud data. When searching for data points in the rectangular filtering window in step 1.1.2, using the index file can greatly improve the calculation efficiency.

Step 1.1.2, traversing the driving track nodes, building a rectangular filtering window in turn, and searching for all data points falling in the rectangular filtering window. The specific search method is:

First, determine the grid unit grid _ij in the grid space constructed by the driving track node route _k in step 1.1.1. Assuming that there are p driving trajectory nodes corresponding to the original vehicle-mounted laser scanning point cloud data, the value range of k is 1, 2, 3, ..., p, and k represents the number of the currently calculated driving trajectory node.

和

的距离d_OA和d_OB，如果d_OA小于width并且d_OB小于length，则该点即落在矩形滤波窗口内。遍历完成，即可确定落在矩形滤波窗口中的所有数据点。Then, in order to speed up the search efficiency, read the grid file generated in step 1.1.1, search the grid unit grid _ij and the eight-neighborhood space of grid _ij , refer to Figure 7, that is, grid _{(i-1)(j- 1)} ，grid _(i-1)j ，grid _(i-1)(j+1) ，grid _i(j-1) ，grid _i(j+1) ，grid _(i+1)(j-1) , grid _(i+1)j , grid _(i+1)(j+1) . For the data points in the above nine grid cells, calculate each data point to

and

The distances d _OA and d _OB , if d _OA is less than width and d _OB is less than length, then the point falls within the rectangular filter window. After the traversal is completed, all data points falling in the rectangular filtering window can be determined.

Step 1.1.3, performing histogram statistical analysis on the elevation values of all data points in the rectangular filtering window obtained in step 1.1.2. The specific statistical method is: take the elevation of the lowest data point in the current rectangular filtering window as the starting point, and partition according to the length of the preset interval. Those skilled in the art can preset the value by themselves. In the embodiment, 0.1 meter is used as the interval length, and there are n intervals in total. Count the number of points falling in each interval in turn until the maximum elevation value in the rectangular filtering window is covered. The n intervals are numbered as 0, 1, 2, ..., n-1 from low to high. Select the 5 intervals with the most points, and determine the interval numbers corresponding to these 5 intervals. Arrange the 5 interval numbers in ascending order, take the interval MaxBin with the most points as the center, traverse the remaining 4 intervals up and down, and select the intervals adjacent to MaxBin in the remaining 4 intervals, That is, the interval adjacent to the elevation of the interval with the most points forms a continuous interval segment. For example, the 5 intervals with the most points are numbered 0, 1, 2, 3, and 5 in ascending order, among which the interval number MaxBin with the most points is 2, and traversing upwards 1, 0, 0, 1, and 2 are adjacent in sequence. Traversing down 3, 5, 2, 3 are adjacent, but 3, 5 are not adjacent; then 0, 1, 3 are all intervals adjacent to MaxBin in turn. After the selection is completed, check whether the minimum interval number in the continuous interval is 0: if it is 0, use the continuous interval as the road surface threshold range, and consider the data points falling in the continuous interval as road surface data point (that is, the road surface point); if it is not 0, the continuous interval segment and all interval segments lower than the continuous interval segment are used as the road surface threshold range, that is, except for the data points in the continuous interval segment that are recognized as roads In addition to the surface data points, all data points below the continuous interval are identified as road surface data points, and the preliminary road surface point cloud data is obtained.

Step 1.2, analyze the preliminary road surface point cloud data obtained in step 1.1 from the two dimensions of distance and angle, arrange them according to the scanning line arrangement, and mark each road surface data point in the preliminary road surface point cloud data The scanline number it belongs to.

In the file storing point cloud data of on-board laser scanning, the point records are generally arranged according to the collection time of the on-board laser scanning system. Read the point records in the point cloud data file in sequence, and calculate the distance and angle respectively. Here, "distance" refers to calculating the distance between each point in the preliminary road surface point cloud data and the next adjacent point; "angle" refers to calculating the distance between each point in the preliminary road surface point cloud data and its two adjacent points before and after Constitute a vector and calculate the angle angle between the two vectors.

Set the distance threshold and angle threshold respectively. If a certain point satisfies that the distance is greater than the set distance threshold and the angle is smaller than the set angle threshold, the point is recognized as the end point of the scan line. During specific implementation, those skilled in the art can preset the distance threshold and angle threshold according to specific road conditions. In this embodiment, the distance threshold is set to 5 meters, and the angle threshold is set to 90 degrees.

Step 1.3, for the point cloud data of the road surface obtained after the scan line sorting in step 1.2, analyze scan line by scan line according to the sequence of scan line numbers. For each scanning line, calculate the elevation difference between each road surface data point in the road surface scanning line and the two adjacent road surface data points in turn, and set the threshold value. If the two elevation differences exceed the threshold value, the other As "noise", remove it. During specific implementation, those skilled in the art can preset the elevation difference threshold by themselves. In this embodiment, the elevation difference threshold is set to 0.2 meters. For the endpoint of the scanning line, it is only necessary to judge the elevation difference between the endpoint and its adjacent point. If it is greater than 0.2 meters, the endpoint is considered to be a "noise point" and removed. After eliminating the noise points from the preliminary road surface point cloud data obtained in step 1.1, the remaining road surface data points are the final road surface point cloud data. After the final road surface point cloud data is removed from the original vehicle laser scanning point cloud data, other ground object point cloud data are obtained.

Further as a preferred embodiment, with reference to Figure 5, the step 2 includes:

Step 2.1, refer to Figure 6, take each driving track node as the center point, set a small radius, construct a circular area, search for the road surface point cloud data falling in the circular area according to the results obtained in step 1.3, and calculate in turn The average elevation value of all road surface data points in each circular area is taken as the road surface elevation value at the node of the driving trajectory. During specific implementation, the value of the radius can be preset by those skilled in the art. It is suggested that the number of road surface data points in the circular area constructed with each driving track node as the center be greater than 10 by setting the value. In this embodiment, the value of the radius is 1 meter.

Step 2.2, referring to Figure 6, traverse the point cloud data of other features obtained after eliminating the point cloud data of the road surface in step 1, and calculate the horizontal distance between each data point in the point cloud data of other features and all the nodes of the driving track in turn . After comparison, select the driving trajectory node closest to the currently traversed data point, and make a difference between the elevation value of the currently traversed data point and the elevation value of the road surface near the driving trajectory node calculated in step 2.1, and realize the elevation correction. The relative elevation value of each data point in other ground object point cloud data relative to its nearby road surface, in order to avoid the classification error and influence caused by the undulation of the road surface.

Step 2.3, referring to Figure 6, construct a rectangular regular grid on the horizontal plane according to the preset sampling interval, project the point cloud data of other features obtained after the elevation correction in step 2.2 into the regular rectangular grid on the horizontal plane, and use the point cloud data of other features The relative elevation value of each data point in the data relative to the nearby road surface counts the maximum elevation value of all data points falling in each grid cell, and takes it as the feature value of the grid cell, thereby generating a point cloud feature image. Constructing a rectangular regular grid on the horizontal plane is a prior art, and the horizontal plane refers to the XOY plane of the geodetic coordinate system, which will not be described in detail in the present invention. In specific implementation, it is recommended that the sampling interval be set to a smaller value, such as 0.1 meters, to avoid dividing the point cloud data belonging to different ground objects into the same grid unit. Through the improved maximum height value method provided in steps 2.1 to 2.3, generate point cloud feature images.

Step 2.4: Preliminary classification of other ground features by threshold segmentation: refer to Figure 6. During specific implementation, the elevation thresholds can be preset for different types of ground features in a specific urban environment, such as buildings, poles, and trees. Each type of ground object is numbered, assuming that there are t types of ground objects, 0 represents the grid unit without point cloud data distribution, and the numbers are 0, 1, 2, ..., t in order from low to high. When running this step, the point cloud feature image generated in step 2.3 is preliminarily classified by threshold segmentation in units of each grid unit, and divided into buildings, poles, trees and other ground objects, etc., and compared with The number corresponding to the category is assigned to each grid unit, and the preliminary classification results of other ground objects are obtained.

Step 2.5, for the preliminary classification results of other ground features obtained in step 2.4, use the principle of cellular automata to perform a certain number of evolutions, and each evolution is for all cells in the cellular space according to the set cellular automaton The rule is traversed once to correct the misclassified grid cell value, improve the accuracy of the classification result, and reduce the impact that may be caused by the improper setting of the sampling interval in step 2.3. For the situation that the ground features are relatively independent of each other and less adjacent to each other, more evolutions can be carried out to ensure sufficient evolution. In view of the situation that there are many adjacent or connected ground features, the evolution times of cellular automata can be determined through experiments: too few evolution times will result in insufficient evolution; too many evolution times will cause excessive evolution . In this embodiment, it is suggested to evolve 50 times to obtain the final classification result of other ground features.

In this embodiment, the construction rules of cellular automata are:

(1) Cell: the grid unit after the preliminary classification result in step 2.4, the state value of each cell is the category number of the corresponding grid unit, and the value range of the cell state value is determined in step 2.4 Feature type numbers 0, 1, 2, ..., t;

(2) Cell space: the collection of all grid cells after the preliminary classification results in step 2.4;

(3) Neighborhood: Moore-type neighbor model, that is, eight-neighborhood neighbor model;

In this embodiment, the cellular automata rules are specifically:

Step a, judge whether the state value S of the current cell is 0, if it is 0, do not make subsequent judgments and keep the state value, if it is not 0, continue to the subsequent step b;

Step b, count the number sum _type of all state values in the Moore neighbors of the current cell, if the number sum _k of a certain state value k is greater than 4, change the state value of the current cell to k and end the The discrimination of the cell, if it does not exist, continue to the subsequent step c;

Step c, count all state values in the Moore neighbors of the current cell to obtain the maximum state value S _max , if S _max is greater than the state value S of the current cell, change the state value of the current cell to S _max , otherwise keep it unchanged.

The specific embodiments described above further describe the technical solutions of the present invention in detail, and are only used to illustrate specific implementation cases of the present invention, and are not intended to limit the scope of the present invention. All equivalent deformations, substitutions or modifications made by those skilled in the art without departing from the spirit and principles indicated by the present invention are still included within the scope defined by the claims of the present invention.

Claims (3)

1. a method for classifying urban features of vehicle-mounted laser scanning point cloud data, is characterized in that, comprises the following steps: Step 1, combined with the driving trajectory data, perform road surface filtering on the original vehicle-mounted laser scanning point cloud data, and separate the road surface point cloud data from other object point cloud data, including the following sub-steps, Step 1.1, according to the driving trajectory data, along the direction of the driving trajectory, with each node of the driving trajectory as the center point, construct a rectangular filtering window, perform histogram statistical analysis on the data falling in the rectangular filtering window to determine the threshold range of the road surface, and obtain a preliminary Road surface point cloud data; Step 1.2: Analyze the preliminary road surface point cloud data obtained in step 1.1 from the two dimensions of distance and angle, arrange according to the scanning line arrangement, and mark the location of each road surface data point in the preliminary road surface point cloud data. scan line number; Step 1.3: Analyze the road surface point cloud data obtained after the scanning lines in step 1.2 are scanned line by scan line according to the order of the scan line numbers, including calculating the relationship between each road surface data point in the scan line and its The elevation difference of two adjacent road surface data points, if the two elevation differences exceed the preset elevation difference threshold, then the road surface data point is removed as a noise point; the preliminary road surface point cloud data obtained from step 1.1 is removed After the noise points, the retained road surface data points are used as the final road surface point cloud data, and the final road surface point cloud data is removed from the original vehicle laser scanning point cloud data to obtain other ground object point cloud data; Step 2, further identifying the point cloud data of other features obtained in step 1 as different types of features, including the following sub-steps, Step 2.1, take each driving trajectory node as the center point, construct a circular area according to the preset, search for the road surface point cloud data falling in the circular area according to the results obtained in step 1.3, and calculate all the roads in each circular area in turn The average elevation value of the surface data points is used as the road surface elevation value at the corresponding driving trajectory node; Step 2.2, traverse and calculate the horizontal distance between each data point in the point cloud data of other ground objects and all the driving track nodes, select the driving track node closest to the currently traversed data point, and set the elevation value of the currently traversed data point to Make a difference with the elevation value of the road surface near the corresponding driving track node obtained in step 2.1, and obtain the relative elevation value of each data point in the point cloud data of other features with respect to the nearby road surface; Step 2.3, according to the preset sampling interval, construct a rectangular regular grid on the horizontal plane, project the point cloud data of other ground objects obtained after the elevation correction in step 2.2 into the regular rectangular grid on the horizontal plane, and count in each grid unit The maximum elevation value of all data points is used as the feature value of the grid cell to generate a point cloud feature image; In step 2.4, preliminary classification is carried out through the preset category threshold, and the number corresponding to the category of each grid unit is obtained, which constitutes the preliminary classification result of other ground objects; In step 2.5, the preliminary classification results of other ground objects obtained in step 2.4 are evolved by using the cellular automaton for a preset number of times to obtain the final classification results of other ground features. 2. according to the said vehicle-mounted laser scanning point cloud data urban feature classification method of claim 1, it is characterized in that: step 1.1 comprises the following sub-steps, Step 1.1.1, according to the range covered by the original vehicle-mounted laser scanning point cloud data, construct a two-dimensional spatial regular grid index, calculate the number of the grid unit to which each data point belongs and generate an index file; Step 1.1.2, traversing the driving track nodes, building a rectangular filtering window in turn, and searching for all data points falling in the rectangular filtering window according to the index file; Step 1.1.3, carry out histogram statistical analysis to the elevation values of all data points in the rectangular filtering window obtained in step 1.1.2, the statistical method is as follows, Take the elevation of the lowest data point in the current rectangular filtering window as the starting point, partition according to the length of the preset interval from low to high, set it into n intervals, and count the points falling in each interval in turn, and in order from low to high Number the n intervals as 0, 1, 2, ..., n-1; select the 5 intervals with the most points, and determine the interval numbers corresponding to these 5 intervals; Arrange the 5 interval numbers in ascending order, take the interval MaxBin with the most points as the center, traverse the remaining 4 intervals up and down, and select the intervals adjacent to MaxBin in the remaining 4 intervals, form a continuous interval segment; Check whether the minimum interval number in the continuous interval segment is 0, If it is 0, then the continuous interval segment is used as the road surface threshold range, and the data points falling in the continuous interval segment are identified as road surface data points to obtain preliminary road surface point cloud data; If it is not 0, the continuous interval segment and all interval segments lower than the continuous interval segment are regarded as the road surface threshold range, and the data points in the continuous interval segment and all data points lower than the continuous interval segment are considered is the road surface data point, and the preliminary road surface point cloud data is obtained. 3. according to claim 1 or 2, the urban feature classification method of vehicle-mounted laser scanning point cloud data is characterized in that: the construction rule of cellular automaton is in the step 2.5, and the cell is the lattice after the preliminary classification of step 2.4 The grid unit, the state value of each cell is the category number of the grid unit, and the cell space is the set formed by all the grid units after the preliminary classification in step 2.4; Each evolution traverses all the cells in the cellular space once according to the set cellular automata rules, and the cellular automata rules include the following sub-steps, Step a, judge whether the state value S of the current cell is 0, if it is 0, do not make subsequent judgments and keep the state value, if it is not 0, continue to the subsequent step b; Step b, count the number sum _type of all state values in the Moore neighbors of the current cell, if the number sum _k of a certain state value k is greater than 4, change the state value of the current cell to k and end the The discrimination of the cell, if it does not exist, continue to the subsequent step c; Step c, count all state values in the Moore neighbors of the current cell to obtain the maximum state value S _max , if S _max is greater than the state value S of the current cell, change the state value of the current cell to S _max , otherwise keep it unchanged.

Images